High-order Newton-penalty algorithms (Q557745)

The author borrows ideas from the interior point algorithms to improve significantly the asymptotic convergence properties of the classical penalty algorithm for equality constrained differentiable nonlinear programs. The classical equality constrained program is solved using the simple quadratic loss penalty function/algorithm. The author explores higher order extrapolations, especially higher order Newton-like methods. First higher order variants of the Newton-Raphson method applied to systems of non-linear equations are considered. The author also obtains improved asymptotic convergence results for the quadratic loss penalty algorithm by using high order extrapolation steps.